Three Sources of Increasing Returns to Scale

Jinill Kim First draft: March 1996 This draft: April 3, 1997
Abstract

This paper reviews various types of increasing returns from a critical perspective. Increasing returns have been introduced both at the rm level and at the aggregate level in a monopolistic-competition model. We show that the degree of the aggregate returns to scale is a linear combination of three return parameters, with the weights determined by the speci cation of a zeropro t condition. Identi cation issues are discussed with an emphasis on recent macro literature. We argue that disaggregate data give information on the market structure rather than the technology. Welfare implications explain why it is important to identify various increasing returns.
Key words :

ety

Increasing Returns; Monopolistic Competition; Returns to VariE32

JEL classi cation :

Federal Reserve Board, Mail Stop 61, Washington, D.C. 20551. Telephone: (202) 452-2715. E-mail: jkim@frb.gov. This is a revised version of a chapter in my dissertation at Yale University. Special thanks to Christopher Sims for his guidance and support. Thanks also to William Brainard, John Fernald, Robert Shiller, Steve Sumner, Michael Woodford, and seminar participants at the Universities of Maryland and Virginia, and Federal Reserve Board for their valuable comments. This paper represents the view of the author and should not be interpreted as re ecting the views of the Board of Governors of the Federal Reserve System or other members of its sta .

1

1 Introduction 2 The Model
2.1 2.2 2.3 2.4

Contents

Firms . . . . . . . Aggregation . . . . Returns to Scale . A Dynamic Model

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

5 8 10 15

2 4

3 Implications 4 A B C D

3.1 Identi cation with Aggregate Data . . . . . . . . . . . . . . . . . . . 19 3.2 Interpretation of Disaggregate Data . . . . . . . . . . . . . . . . . . . 21 3.3 Comparison with a Social Planner . . . . . . . . . . . . . . . . . . . . 24

19

Further research External Increasing Returns Input Fixed Cost Cost Minimization Fixed Cost Externalities

26 28 29 31 33

1 Introduction
The hypothesis of noncompetitive markets and/or increasing returns to scale has recently been used in dynamic stochastic general-equilibrium (DSGE), more often called real-business-cycle, models. Using the Solow residual as a measure of productivity changes is appropriate only under the joint hypothesis of perfectly competitive markets and constant returns to scale. In a series of papers, Hall (1986, 1988, 1990) argues that evidence from the Solow residual is not consistent with this maintained hypothesis but with the alternative hypothesis of noncompetitive markets and/or increasing returns to scale.1 Under this alternative hypothesis, the Solow residual has
Imperfect competition makes equilibrium possible in the presence of increasing returns. Increasing returns are compatible with competitive rms if the increasing returns are external to the rms. Internal returns may be motivated as a representation of external ones, as in Beaudry and Devereux (1995a). The two types are compared using models with both types of increasing returns in Appendix A.
1

2

endogenous components which cause it to over-represent the contribution of productivity shocks. Furthermore, this alternative hypothesis helps explain some puzzles in the DSGE literature, e.g. little correlation between employment and productivity. Following Dixit and Stiglitz (1977) and Blanchard and Kiyotaki (1987), the monopolistic-competition framework has been widely used in macroeconomics. The assumption of unrestricted entry and exit implies that pro ts are zero in equilibrium.2 In a monopolistically competitive market, the technology of constant returns to scale lets rms produce positive pro ts regardless of their size. Introducing increasing returns at the rm level leaves room for reducing pro ts to zero.3 The objective of this paper is to discuss three di erent types of increasing returns in a monopolistic-competition model and to derive implications for the related literature. There are two ways of introducing increasing returns at the rm level. The more conventional way is including xed costs as part of a rm's technology. This way has been followed whenever a zero-pro t condition is imposed. An alternate way is amplifying the constant-returns-to-scale term by a power larger than one, which amounts to diminishing marginal cost. When we incorporate both sources of increasing returns simultaneously, as in Hornstein (1993), their e ect on the aggregate returns to scale is di erent from each other. Increasing returns due to the third source occurs only at the aggregate level. It involves a technology or a preference for the variety of goods. The introduction of a new good might enhance the production e ciency and the consumption convenience. Romer (1987) focuses on this as an engine of growth and Matsuyama (1995) relates this to complementarities and cumulative processes of macroeconomics. The model in Devereux, Head and Lapham (1996a), even without productivity shocks, generates business cycle uctuations of real variables from government spending shocks since these a ect the variety of goods. This paper shows that, in a static model, the resulting degree of aggregate returns to scale is the average of the second and the third sources of increasing returns, without any in uence of positive xed costs. The derivation of aggregate returns from a rm's technology involves two steps. First, the di erentiated outputs are aggregated to produce a measure of aggregate output. Second, a zero-pro t conSee Benassy (1991) for quali cations of zero-pro t conditions. The assumption of zero pro ts matches the observation in Hall (1990) and Rotemberg and Woodford (1995) that there are no signi cant pure pro ts in the United States. 3 Not that all papers in DSGE literature impose a zero-pro t condition. Hairault and Portier (1993) and Beaudry and Devereux (1995b) do not impose a zero-pro t condition and so parameterize both xed cost and the number of rms. In such models, the rm-level returns to scale are the aggregate returns to scale and the permanent presence of positive pro ts remain unexplained. For example, the steady-state pro t rate is 17% in the benchmark model of Hairault and Portier (1993).
2

3

dition is imposed upon the aggregate version of a rm's technology. Speci cation of a zero-pro t condition determines the weights of the averaging. In a dynamic model where adjustments to zero pro t are not instantaneous, the market structure of monopolistic competition plays a role|the slower the adjustments, the larger the role. Even if market structure does not directly a ect the technology, this source in uences the response of output in a way indistinguishable from the previous two ways. The aggregate dynamics of a model which combines various sources of increasing returns to scale show that there are identi cation problems in the recent macroeconomics literature using the framework of monopolistic competition. We compare various papers to see how they specify a zero-pro t condition and what the resulting degree of returns to scale is. We also argue that the literature using disaggregate data provides information di erent from what it intends to provide: on the market structure rather the technology. Lastly, welfare implications are discussed from the perspective of a social planner who does not need to satisfy zero-pro t conditions. While having similar positive implications for the aggregate returns, various increasing returns have di erent normative implications.

2 The Model
To illustrate the points in as simple a structure as possible, we analyze only the production side of the economy. This analysis is tractable and gives much insight on how di erent returns to scale interact with one another. Most papers on monopolistic competition deal with both the production and the consumption side of an economy. However, introducing a utility function complicates the model so that it is di cult to disentangle the production features from consumer behavior. Our model is a partial-equilibrium model, since the production side generates the demand for inputs. The transformation of this model into a general-equilibrium framework is straightforward by stacking it with a consumer problem and, if needed, a government problem. The consumer problem would generate the supply function of aggregate inputs through a labor-leisure choice and capital accumulation. Therefore, throughout this paper, we may consider the aggregate inputs as exogenous variables. Since a zero-pro t condition is crucial in deriving the economy-wide returns to scale, we will be very careful in discriminating two meanings of `production function.' A structural production function is a purely technological relation without reference to the equilibrium condition of zero pro ts. However, a production function in a reduced form, whether a rm's or an aggregate one, is a combination of the appropriate technology and a zero-pro t condition. That is, a reduced-form 4

production function is a structural production function with a zero-pro t condition imposed. Now our objective of this section is to transform a rm-level structural production function, Eq. (1), into an aggregate reduced-form production function, e.g. Eqs. (16) and (24). This transformation is a contribution to the literature since it simpli es one step of complex DSGE models and so makes it easy to understand their production features. We start with a static model since it is a special case of a dynamic model. The static model serves as a steady-state, or low-frequency in general, feature of the dynamic model. As a preparation for the analysis of the aggregate variables, we analyze the behavior of rms. Firms are identical except for the heterogeneity of outputs. Firm i produces yi units of net output under a technology of increasing returns to scale:

2.1 Firms

yi = Ai ki i li1

i

i

i

;

(1)

with the restrictions that 0 i 1, i > 0, and i > 0. Ai denotes the productivity shock, ki is the capital stock and li is the quantity of labor. Since productivity shocks are not crucial in deriving the implications on returns to scale, they are normalized to 1 except when necessary for discussing econometric issues. The parameter i represents what rm i should pay at each period regardless of its activity level. For example, a rm advertises its good each period to maintain its market share. Note that the xed cost is measured in units of its own output, not its inputs.4 Additionally, this paper follows the convention that a rm's xed cost is exogenous to the rm. The presence of a xed cost makes it possible to impose a zero-pro t condition, as in Hornstein (1993), Rotemberg and Woodford (1995), Beaudry and Devereux (1995a), and Devereux, Head and Lapham (1996a,b), and is a source of increasing returns to a rm's technology. However, in a static model, rm-level increasing returns due to xed costs are not transmitted to the increasing returns of an aggregate production function in a reduced form, Eq. (16). If i is greater than 1, a rm's gross output features additional increasing returns to scale. This source of increasing returns has not been popular in the literature. Actually, most papers in the DSGE literature restrict i to be exactly equal to 1, except for Hornstein (1993) and Benhabib and Farmer (1994). This paper names
Chatterjee and Cooper (1993) and Yun (1996) assume that xed costs are a part of inputs. The models with input xed costs behave in a similar way. See Appendix B for a model with xed costs as part of its inputs.
4

5

this source `diminishing marginal cost.' Appendix C on cost minimization shows that the bigger i is, the smaller the slope of marginal cost is. Hornstein (1993) call this `the scale coe cient.' However, we will show that the degree of returns to scale depends also on other parameters. Furthermore, diminishing marginal cost may not a ect the response of aggregate output, depending on the speci cation of a zero-pro t condition. These two sources of increasing returns to scale at the rm level are added up in overall returns to scale of a rm's technology. From the perspective of a rm, to whom the xed cost is exogenous, the log-linearized reduced-form production function is: h y+ i That is, the degree of returns perceived by a rm is the product of the degree of diminishing marginal cost and the ratio between the gross and the net output. The ratio turns out to be i , where is the degree of market power. This degree is assumed to be greater than i and de ned in Eqs. (5) and (6). So the degree of returns to scale of a rm's technology is equal to the degree of market power. However, this exercise is meaningless in that we have not incorporated a zero-pro t condition into a rm's technology. Actually, we will see that the presence of xed costs by itself does not imply increasing returns in a static model. However, the degree of market power turns out be the right measure in a dynamic case where there are only steady-state adjustments to zero pro t, without any period-by-period adjustments. A rm maximizes its pro t:5
i

yi ' ki i li1

i

i iy i i

:

= piyi PZki PWli;

where pi is the price of rm i's product, P is the general price level, Z is the real rental rate of capital, and W is the real wage. In models with monopolistic competition, the price of outputs, pi , depends on the amount produced, yi. The inverse demand for the output of the ith rm will be derived later in Eq. (8), which involves the parameter representing the market structure of monopolistic competition, . The assumption that rms rent capital rather than holding and accumulating it simpli es the analysis. Under this assumption, we do not need to explain where entering rms buy capital or where exiting rms sell remaining
An alternative, and probably more conventional way of analyzing the rm behavior is the framework of cost minimization in Appendix C. However, the pro t-maximization framework is better for understanding the mechanics of the zero-pro t condition since it expresses the endogenous variables as a function of aggregate inputs, considered exogenous in this paper. Furthermore, extension to a dynamic model becomes more complicated in the framework of cost minimization.
5

6

capital. Capital and labor are homogeneous and a rm behaves as a price-taker in the input markets. Noting that yi is a function of (ki; li), the rst order conditions with respect to ki and li are:6

! ! yi + i pi = PZ; i i ki ! ! yi + i pi = PW: i (1 i) li

(2) (3)

Input prices determine the ratio between gross output |not net output| and each input. This point will be important later in comparing the dynamics of output with those of input prices. The rst order conditions, Eqs. (2) and (3), determine the optimal capital and labor, depending on the demand curve and the size of the xed cost.7 A rm's maximized pro t is:
i

= pi 1

"

i

!

ki

i

l1
i

i

i

#

i

:

This derivation shows that neither form of increasing returns to scale is allowed in a perfectly competitive economy, i.e. when = 1. Introducing increasing returns would generate corner solutions. Therefore, the discussion of increasing returns in a competitive economy in Hall (1988) is futile.8 It is also interesting to note that marginal costs decreasing at a faster rate do not always bene t rms. Other things being equal, a rm would produce more output, which would decrease its pro t by lowering its output price. If ki i li1 i is below a certain level, a faster rate means lower pro ts. Otherwise, pro ts become higher as i increases only when i is close to 1. The zero-pro t condition which should hold in equilibrium implies the following restriction: ! i (4) ki i li1 i i : i= 1 The assumption of positive xed costs is equivalent to the assumption that the degree of diminishing marginal cost ( i) may not be larger than the degree of market
Here we neglect the price-index e ect of individual pricing decisions. See d'Aspremont, Ferreira and Gerard-Varet (1996) for this issue. 7 If marginal cost is decreasing, the solution to (2) and (3) might not be a pro t-maximizing choice. The restriction that i is nonnegative is su cient to guarantee that the choice de ned by (2) and (3) maximizes pro ts in equilibrium. 8 This is corrected in Hall (1990), where rms have market power under increasing returns.
6

7

power ( ). This equation may be interpreted as determining the size of the xed cost. However, it may also be interpreted as a condition deciding the number of rms, since ki and li depend on the number of rms. Since aggregate inputs are considered exogenous in this paper, we should transform the zero-pro t condition into a relation among aggregate variables. Before discussing aggregate implications of two rm-level increasing returns, we need to specify how the economy values the variety of goods in aggregation, which introduces the third source of increasing returns. From a macroeconomic viewpoint, heterogeneous outputs need to be aggregated. A convention is introducing an additional agent in the economy, called an aggregator. The aggregator is equivalent to a rm producing a nal good. The presence of the aggregator can be avoided, when every agent chooses the goods and labor index composition optimally, as in Blanchard and Kiyotaki (1987) and Hairault and Portier (1993). In this case, aggregation is a matter of preference as well as technology. Since this choice is static and the same for all agents, notation is simpli ed when the aggregator solves the problem instead. The aggregator purchases di erentiated outputs from rms, which are described by an N -dimensional vector, (y1; y2; : : : ; yN ), and transforms them into Y units of a nal good. In the literature on monopolistic competition, the aggregating function follows both constant returns to scale and constant elasticity of substitution. The speci cation used by Dixit and Stiglitz (1977), and thenceforth conventional, is:
N X 1! Y= yi ; i=1

2.2 Aggregation

(5)

with > 1. The elasticity of substitution between two di erentiated outputs is constant at 1 . Since is the markup in equilibrium, it is de ned as the degree of market power. Besides determining the elasticity of substitution, the parameter plays an additional role involving the variety of goods. If all goods are hired in the same quantity, y, then aggregate output is:

Y = N y:
Thus, there are increasing returns to variety (N ), together with constant returns to quantity (y). If the number of rms is constant in the model, this type of increasing returns does not produce aggregate increasing returns. Hornstein (1993) and Beaudry and Devereux (1995a) fall into this category. However, in a model 8

where the number of rms is endogenous, increasing returns to variety do generate aggregate increasing returns, as in Devereux, Head and Lapham (1996a,b). They argue that increasing returns to variety capture the spirit of a thick-market e ect and also explain the procyclical behavior of job creation and the entry of new rms.9 In the conventional speci cation of Eq. (5), the degree of increasing returns to variety is linked to the degree of market power. Theoretically speaking, there is no a-priori reason to prefer the presence or the absence of increasing returns to variety. Furthermore, with all the agreement on the presence of increasing returns to variety, the degree does not need to be related to the degree of market power, as in Eq. (5). This paper parameterizes, and so disentangles, the degree of returns to variety separately from that of market power. Dixit and Stiglitz (1975)|a workingpaper version of Dixit and Stiglitz (1977)| and Benassy (1996) also disentangle the two parameters, but this is not related to the degree of returns to scale. Disentangling the two parameters is not just a theoretical curiosity. We will show that input prices, as a function of aggregate variables in Eqs. (26) and (27), are a function of the variety parameter but not of the market structure parameter. Therefore, only the former contributes to the existence of equilibrium indeterminacy. Besides, comparing the economy of monopolistic competition with that of a social planner, we will show that the two parameters have di erent welfare implications. These two points will be discussed after the algebra of the model. This paper parameterizes the aggregator as follows:10

Y =N

N X 1! yi ; i=1

(6) (7)

which results in the following equation for aggregate output at symmetry:

Y = N y:

The new parameter, , represents the returns to variety. If it is greater than 1, there are increasing returns to variety. Note that Eq. (5) corresponds to the case when is equal to . If it is equal to 1; the aggregator implies constant returns to variety
Romer (1987) and Devereux, Head and Lapham (1996a,b) call this a return to `specialization' rather than `variety.' Following the title of Dixit and Stiglitz (1977), Chatterjee and Cooper (1993) use the terminology of `product diversity.' 10 A new multiplicative term, N , is called the \public-good feature of diversity" in Dixit and P P Stiglitz (1975). Two equivalent representations are N N ( 1) yi and N (N yi ) . i=1 i=1 This function is discontinuous with respect to the introduction of a new good. However, the discontinuity disappears if we assume a continuous goods space.
9
1 1

9

as assumed in Rotemberg and Woodford (1995). They motivate constant returns as a normalization, but this assumption is a restriction rather than a normalization. The aggregator is assumed to maximize its pro t: = PY
N X i=1

pi y i ;

where all price variables are exogenous to the aggregator and the aggregate price index is de ned as:11 ! ( 1) N 1 1 ( ) Xp P =N : i The rst order condition with respect to yi reduces to a constant-elasticity inverse demand function: 1 ( ) yi : (8) pi = PN Note that the new parameter, ; a ects only the level of the demand without a ecting its elasticity. Hence the rst order conditions of the rms, Eqs. (2) and (3), are not sensitive to the parameterization of returns to variety. Besides aggregating di erentiated goods, macroeconomics has also made it a convention to scrutinize a symmetric equilibrium under identical technologies of the rms. The homogeneity of capital and labor implies the following relations in equilibrium: K = Nk; L = Nl; (9) where K and L denote aggregate capital and aggregate labor, respectively.
i=1

Y

2.3 Returns to Scale

Based upon previous discussions on rms' behavior and aggregation, we now derive the aggregate reduced-form production function, i.e. an aggregate version of rms' technology with a zero-pro t condition imposed. Aggregation of homogeneous inputs in Eq. (9), together with aggregation of outputs as shown in Eq. (7), transforms the rms' technologies, Eq. (1), as follows:

Y = N(
11
( 1)

)

K L1

N :

(10)

At symmetry, P = N 1 p. Increasing returns to variety imply that an increase in the number of rms leads to a decrease in the price index due to an e ciency gain. See Feenstra (1994) for an application to import goods. Existing price indices do not adjust bene ts to variety: they assume = 1.

10

This is the aggregate structural production function. Both the aggregate gross output and the aggregate xed cost feature an element of returns to variety due to aggregation. Note that we have not yet imposed a zero-pro t condition, which is a part of the equilibrium conditions. That is, the aggregate production function in a reduced form is derived only after the zero-pro t condition is imposed. We now derive what the zero-pro t condition, Eq. (4), implies for the aggregate reduced-form production function. The relationship between individual and aggregate inputs transforms the zero-pro t condition as follows:

N = 1

!

K L1

:

(11)

Note that this does not involve the degree of returns to variety, . This aggregate version of the zero-pro t condition has been interpreted in two di erent ways in the literature. The rst interpretation endogenizes the number of rms, N , as a function of the aggregate variables as follows:

0 1 N =@

11 A K L1 :

(12)

If there is a change in the economy causing a positive pro t, e.g. a positive productivity shock, the number of rms increases to take advantage of this change, thus reducing pro ts to zero. The xed cost of a rm is assumed to remain constant. Rotemberg and Woodford (1995) and Devereux, Head and Lapham (1996a,b) adopt this interpretation. This interpretation has a conceptual problem associated with the de nition of the equilibrium, since an equilibrium with entry and exit involves a change of the goods space.12 Conversely, the second interpretation considers the xed cost as an endogenous variable, that is: ! =N 1 K L1 : (13) Taken literally, a rm's xed cost increases with aggregate inputs and is also a ected by the market structure represented by the degree of market power, . This interpretation has been less attractive than the former, since the amount of the xed cost is not a technological parameter but a function of other parameters and exogenous variables.13
12 13

See Benassy (1991) for the details. The endogeneity with respect to the degree of market power might be rationalized by the argument that a larger xed cost (e.g. advertising) should be paid at each period in an economy with a larger degree of market power.

11

However, an increase in the xed cost per rm is equivalent to job creation on the intensive margin. That is, output per rm is proportional to the xed cost. Furthermore, over low frequencies of a growing economy, it is reasonable to assume that xed costs grow as rm size grows over time. No papers have argued for this interpretation seriously, but some follow this interpretation implicitly by assuming that the number of rms is constant. If the number of rms is assumed to be constant, as in Hornstein (1993), Yun (1996), and Beaudry and Devereux (1995a), the only way to achieve the zero-pro t condition is by endogenizing the xed cost of a rm. As with the speci cation of the aggregator, there is no a priori reason to prefer either interpretation. In general, both the xed cost and the number of rms may change. This paper, rst in the literature, incorporates this generality by de ning an additional parameter, " 2 0; 1], which represents the ratio of the intensive and extensive margins. Our parameterization postulates the xed cost and the number of rms as follows:14 #" " ! 1 1 = K L1 ; (14)

N =

"

1

!

K

L1

#1

"

;

(15)

where is an arbitrary constant. This constant a ects only the level of variables, but not their percentage deviation. We can interpret the parameter " as one related with the elasticity of supplying new rms. Manipulating Eqs. (14) and (15), we have a constant-elasticity supply schedule,

N "=

1 ";

where is interpreted as the marginal cost of some specialized resource, e.g. entrepreneurial ability, required to create a new rm. For example, when " = 0, the supply of entrepreneurial ability is in nitely elastic at a price of per unit. Note that the extreme cases of " = 0 and " = 1 correspond to the two interpretations in the literature, Eqs. (12) and (13). This paper shows that our generalization is analytically tractable and gives intuition for the interaction among the various increasing returns. Before discussing the general case containing the parameter ", we consider the two extreme cases.
This parameterization of endogenous xed costs can be reconciled with that of exogenous xed costs, when negative externalities are introduced via xed costs as in Appendix D. Optimal decision of xed cost in industrial-organization literature endogenizes the " parameter, but constant " is not a bad approximation when time unit if constant.
14

12

Suppose that the number of rms is endogenous, i.e. " = 0, then the aggregate output in equilibrium is derived by substituting Eq. (12) into Eq. (10): !( 1) ( 1) 1 Y = K L1 : In this case, , the degree of returns to variety, determines the degree of aggregate returns to scale. The intuition behind this is that the rm size is constant while the number of rms is variable. Since the endogenous number of rms is proportional to the constant-returns-to-scale term (K L1 ) in Eq. (12), the channel of returns to variety works while overriding that of diminishing marginal cost. Note that the economy may produce more output with a larger xed cost. If is smaller than , output increases as the xed cost increases. This apparently abnormal result comes from the following intuition: since the degree of diminishing marginal cost is higher than that of returns to variety, it is better to have larger rms. The gains from larger rms o set the losses from a larger aggregate xed cost. However, if we assume an endogenous xed cost, i.e. " = 1, then the aggregate output behaves di erently. Using Eq. (13), the aggregate technology of Eq. (10) is transformed as follows: YN = N ( ) K L1 : Note that the degree of returns to scale is , the degree of diminishing marginal cost. Opposite from the previous case of an endogenous number of rms, the channel of diminishing marginal cost is in e ect while there is no e ect of the returns to variety. In a sense, this is trivial since the number of rms is constant. A rm's output and xed cost move proportionally with the aggregate output. Inputs per rm move also proportionally with the aggregate inputs. Note that an increase in the number of rms may increase or decrease the aggregate output, depending on the sign of ( ). If the degree of returns to variety is higher than that of diminishing marginal cost, it is better to have more rms in the economy. Now let's consider our general case where both the xed cost and the number of rms change according to Eqs. (14) and (15). It is natural to speculate that the degree of returns to scale is between the two extreme cases. Substituting Eqs. (14) and (15) into Eq. (10), the aggregate production function in a reduced form is simpli ed as follows: !(1 ")( 1) ( 1) 1 K L1 " +(1 ") : (16) Y = The degree of returns to scale in this general case is " + (1 ") ], a convex combination of and . The degree of returns to scale in the aggregate economy depends 13

on the choice of three parameters representing diminishing marginal cost ( ), returns to variety ( ), and the ratio of the two endogenous changes ("). Note that increasing returns at a rm level due to positive xed costs does not generate increasing returns at an aggregate level. The e ect of monopolistic competition on the aggregate output is now clear. Both the xed cost and the number of rms are an increasing function of the degree of market power, . Since a larger degree of market power means a larger pro t, we need more rms and a larger xed cost per rm to reduce the pro t to zero. Given a certain amount of the aggregate inputs, a higher degree of market power decreases the amount of di erentiated output due to an increase in the xed cost and a decrease in inputs per rm. However, if the number of rms is endogenous (" < 1) and the degree of returns to variety is larger than that of diminishing marginal cost ( > ), an o setting mechanism is at work, since a higher degree of market power increases the number of rms.15 This mechanism overturns the decrease in a rm's output if the degree of market power is lower than the degree of aggregate returns to scale, " + (1 ") ]. In other words, an economy with monopolistic competition would produce maximum output T the degree of market power were if equal to the degree of returns to scale. The -shaped graph in the rst plot of Figure 1 on page 27, denoted as Ym, shows this property. For example, in a model with a Dixit-Stiglitz aggregator ( = ) and an endogenous number of rms (" = 0), the monopolistic-competition economy is optimal in the sense that it produces the most aggregate output. This optimality pertains to the comparison among di erent market economies and has nothing to do with the comparison with a social planner, which will be discussed later. Before moving on to a dynamic model, we analyze the mechanics of input prices. The pro t maximizations of the rms have implications for the rental rate and the wage, Eqs. (2) and (3), which result in the following aggregate relations:

Z =

W = (1

Y; K

)Y :

(17) (18)

Although the dynamics of input prices depend on gross output, they mimic those of net output because the aggregate xed cost is proportional to the aggregate gross output, and so the net output. However, this property does not extend to a dynamic case. This point turns out to be important in the discussion of identi cation issues.
It is easy to check that the aggregate output is decreasing over the whole range of the degree of market power if both conditions do not hold.
15

L

14

Up to now, neither the size of the xed cost nor the degree of market power a ects the degree of returns to scale. Then, is there no point of introducing monopolistic competition and the xed cost in the behavior of returns to scale? Yes, the presence of the xed cost does a ect returns to scale in a dynamic model where adjustments to zero pro t are not instantaneous. Note that the static case considered above is analogous to the Marshallian long run, in that no specialized resource prevents zero pro ts from being achieved. Following the same spirit, we interpret a dynamic case considered below as the Marshallian short run since existing rms may earn quasi-rents. To consider a dynamic case, the aggregate technology without the zero-pro t condition imposed, Eq. (10), is rewritten with time subscripts for the variables.

2.4 A Dynamic Model

Yt = Nt(

)

Kt L1 t

Nt t :

(19)

This structural production function does not directly involve the degree of market power, . However, this degree representing the market structure may a ect the degree of returns to scale of the aggregate reduced-form production function in a dynamic model. Apart from the choice of endogenous variables in a static model, an additional consideration arises in a dynamic model: how fast does the economy move towards the state of zero pro ts? Two extreme cases have been considered in the literature. In the case of full adjustment, pro t is zero at every period. This speci cation is used in Devereux, Head and Lapham (1996a,b). The short-run dynamics are the same as that of the static model described in Eqs. (16){(18). The other extreme is the case of no short-run adjustment. Pro t is zero only at the steady state, so the short-run dynamics of the xed cost and the number of rms are not a ected even if pro t is not zero in a particular period. Hornstein (1993) and Beaudry and Devereux (1995a) adopt this speci cation. In this case, returns to scale are governed by neither diminishing marginal cost nor returns to variety. They are governed by the degree of market power, which is a function of the degree of substitution among di erentiated goods. These two extreme cases are compared in Chatterjee and Cooper (1993). Another contribution in this paper is to generalize the speci cation of zero pro ts by introducing an additional parameter, 2 0; 1], representing the speed of adjustments.16 This speed is inversely related to cost of adjustments, e.g. entry and exit.
Rotemberg and Woodford (1995) is the only example which does not follow one of the two extreme cases. The calibration corresponds to a very small of this paper. Optimal decision of
16

15

We assume that the xed cost and the number of rms adjust period by period, but not fast enough to achieve zero pro t at every period:17 ~ (1 ) ; t = ~ Nt = N N (1 ) ; ~ where ~; N guarantees zero pro t at every period and ; N guarantees zero pro t at the exogenous steady state. Note that no adjustment corresponds to the case when = 0, and full adjustment, when = 1. Speci cally, the four variables are de ned as follows: ~ = 1 1 ~ " 1 1 = ~ N = ~ 1

"

! !

Kt K Kt K

L1
t

#"~ #"

; ;
" ~ "

(20) (21)

"

! !

L1 L1
t

#1 #1

N =

"

; :

(22) (23)

1

L1

We will see that only ", a parameter regarding zero pro t at every period, matters ~ for the dynamics of the model. The other parameter, ", a ects only the steady-state properties. Alternatively, the speed of adjustments can vary between the xed cost and the number of rms with the assumption that " = . In this case, the three ~ parameters are ("; ; N ). However, this new model is equivalent to our model with the parameters de ned as follows: = " + (1 ") N ; " = "; " = 1 ~ ": 1 Substituting Eqs. (20){(23) into Eq. (19), the aggregate reduced-form production function in the dynamic model is as follows:

Yt =

1

h

Kt Lt1

i (~ +(1 "

") )+(1 ~

)

2

h

Kt L1 t

i (~ +(1 "

") ) ~

;

(24)

entry and exit in industrial-organization literature endogenizes the parameter, but constant is an approximation under constant time unit. A higher frequency would imply a lower . 17 Our speci cation with an additional parameter preserves the simple one-period nature of the model. The model becomes intertemporal if we introduce partial adjustments by assuming the xed cost and the number of rms predetermined, as in Ambler and Cardia (1996).

16

where
1 2

= =

~ ~

1 1

( (

1) 1)

1 1

! !

(1 ")+(1 )(1 ")]( ~
" +(1 ") ~ ~

1) h

K L1

i(1

)(1 ")(

)

; :

+(1 ) "

+(1 ")

h 1 i(1 KL

)(" +(1 ") )

Note that the two exponent terms in Eq. (24) are di erent from each other and that neither of them involves the degree of market power, . Gross output uctuates strictly more than the aggregate xed cost, unless adjustments are instantaneous. This nonlinear reduced-form production function has such complicated dynamics that it cannot be compared with the static case. However, in a linearized version of the dynamic model, the behavior of the variables is comparable to the static model. In practice, most DSGE papers use a linearized version. The log-linearized version of Eq. (24) is: ^ Yt = (~ + (1 ") ) + (1 " ~ ^ ) ] Kt + (1 ^ ) Lt ; (25) where xt is the percentage deviation of xt from its steady state. This clearly shows ^ that market structure a ects output uctuations in a way indistinguishable from the two previous sources of increasing returns, diminishing marginal cost and increasing returns to variety. The degree in the dynamic case is a convex combination of the degree of the static case in Eq. (16), " + (1 ") ], and the degree of market ~ ~ power, . Equivalently, the degree of returns to scale is a convex combination of the three parameters: the degree of diminishing marginal cost ( ), the degree of returns to variety ( ), and the degree of market power ( ). Note that the degree of returns to scale in the long run, i.e. at the steady state, would be " + (1 ") ] which is di erent from that of the short run in Eq. (25). This helps explain the di erence between low- and high-frequency uctuations, both of which are important in macroeconomics. In the extreme case where there is no period-by-period adjustment, the degree of short-run returns to scale is the degree of market power; the other two parameters do not enter at all. The intuition of this extreme case is as follows. Since aggregate xed cost does not respond to the change of aggregate inputs, net aggregate output uctuates more than gross aggregate output. Furthermore, the amount rms produce relative to the size of the xed cost depends on the degree of market power. The result of this extreme case shows that the introduction of diminishing marginal cost in Hornstein (1993) and Beaudry and Devereux (1995a) does not a ect the dynamics of the aggregate reduced-form production function. However, 17

this does not mean that the introduction has no in uence on the dynamics of a general-equilibrium model at all. The degree of diminishing marginal cost a ects the dynamics of input prices as follows. In a dynamic case, the dynamics of the rental rate and the wage are as follows:
" ! hK L1 i (~ +(1 "~) )+(1 ) 1 t t Zt = ; Kt " ! hK L1 i (~ +(1 "~) )+(1 1 t t Wt = (1 ) Lt

(26)
)

:

(27)

Regardless of the assumption about adjustments, the degree of market power does not a ect the dynamics of the input prices. The dynamic properties of input prices are governed by those of gross output rather than net output, since xed costs and the number of rms are exogenous to the rm's input decision. For example, in models where the presence of xed costs is the only source of increasing returns as in Hornstein (1993) and Rotemberg and Woodford (1995), the introduction of monopolistic competition and increasing returns does not a ect the dynamics of input prices as a function of aggregate inputs. Note that the exponent term relevant for the input-price dynamics is strictly smaller than that of the aggregate returns to scale, unless the adjustments are instantaneous.18 This di erence has an implication for deriving indeterminacy from increasing returns to scale, as in Benhabib and Farmer (1994). Since the dynamics of the input prices are critical for the existence of indeterminacy, all sources of increasing returns do not contribute to its existence. Of the three sources, the degree of market power has nothing to do with the existence of indeterminacy. The bottom line of the model is as follows. The degree of returns to scale of the aggregate reduced-form production function is a convex combination of three parameters: the degree of diminishing marginal cost, the degree of returns to variety, and the degree of market power. The weights depend on the speci cation of a zeropro t condition. Unless zero pro t is imposed period by period, the dynamics of input prices have information independent of the aggregate reduced-form production function.
In other words, even if we have the dynamics of output and each input, we do not derive the dynamics of the input prices. The dynamics of the input prices have some independent implications.
18

18

3 Implications
This section derives implications of the model and, based upon them, reviews and critiques existing literature. The model is related to two branches of empirical literature. A direct implication comes from the behavior of the aggregate reducedform production function. The DSGE literature on monopolistic competition uses a speci c parameterization of this paper. Our general model in this paper gives a warning sign to both calibration and estimation approaches. Another implication of the model involves the works using disaggregate data to identify some parameters. This paper gives interpretations for the estimates this literature have found, di erent from how they used to be interpreted. In addition to these empirical implications, this paper derives welfare implications. We compare an economy of monopolistic competition with a social planner's economy and nd that some previous welfare results are due to restricted speci cations. To facilitate the discussion of identi cation issues, we begin by comparing the speci cations of how the existing literature parameterizes the speed of adjustments ( ) and the period-by-period endogeneity between the xed cost and the number of rms (~). " For convenience, recall that the degree of returns to scale of the aggregate reducedform production function is (~ + (1 ") ) + (1 ) ]. Therefore, this value is " ~ of sole importance in determining the aggregate dynamics of the production function. The dynamics of input prices are governed by (~ + (1 ") ) + (1 ) ], " ~ the degree of returns to scale of aggregate gross output. Hornstein (1993) and Beaudry and Devereux (1995a) assume that only the xed cost is endogenous and that the adjustment to zero pro t occurs only at the steady state. Since the resulting degree of aggregate returns is , diminishing marginal cost and increasing returns to variety do not a ect the dynamics of the aggregate reduced-form production function. However, this does not mean that, in a particular general-equilibrium model, the dynamics of output depend only on the degree of market power. Since the dynamics of input prices depend on diminishing marginal cost and returns to variety, they also a ect output in a general-equilibrium model.19 In Rotemberg and Woodford (1995), the endogenous variable is not the xed cost but the number of rms. However, this does not make much di erence because their calibrated speed of adjustments is very low, a small . If the responses of the number
For example, the second and the third cases of Hornstein (1993) do not produce the same results even if the only di erence between the two cases lies in the degree of diminishing marginal cost.
19

3.1 Identi cation with Aggregate Data

19

"n ~

0 1

0 1 Rotemberg and Woodford (1995) Devereux, Head and Lapham (1996a,b,c) Hornstein (1993) Beaudry and Devereux (1995a) Table 1: Examples of Speci cation

of rms happen only at the steady state, a zero , then di erent assumptions on endogeneity do not make any di erence at all. The endogeneity of the number of rms matters if the zero-pro t condition is imposed period by period, as in Devereux, Head and Lapham (1996a,b,c). The dynamics of both output and input prices depend on the degree of returns to variety. The degree of diminishing marginal cost and that of market power do not a ect the properties of the variables, as far as the percentage deviations are concerned. Table 1 summarizes the speci cations. The rst column corresponds to a model where adjustments to zero pro t occur only at the steady state.20 The second column contains the static version of this paper, where returns to variety matters. The rst and the second rows correspond to models of endogenous number of rms and endogenous xed cost, respectively. Discussion of econometric issues such as identi cation naturally involves the speci cation of an error structure. A convention is to interpret the productivity shocks as a random variable. Accordingly, this section restores the notation for productivity shocks, At . Note that identi cation issues can be discussed only in the context of a speci c model and available data. From the calibration point of view, identi cation issues are interpreted as follows. If the model is not identi ed, di erent parameter calibrations may result in the same model. Suppose that our model is the aggregate reduced-form production function with exogenous steady states and that we have aggregate data on output, capital and labor. Augmenting the linearized aggregate reduced-form production function, Eq. (25), with productivity shocks as an error structure, we have: ^ Yt = (~ + (1 ") ) + (1 " ~ ) ] ^ Kt + (1 ^ ^ ) Lt + 1 At :

!

Using the aggregate data, we can draw inferences about the share parameter, , and the aggregate returns to scale, (~ + (1 ") ) + (1 ) ]. The three parameters " ~ representing the degrees are not identi ed separately. From the calibration point
Rotemberg and Woodford (1995) is included in this category since is very small, even if not exactly zero.
20

20

of view, di erent parameter calibrations may result in the same aggregate returns to scale. For example, the dynamic properties of Rotemberg and Woodford (1995) with slow period-by-period adjustments, i.e. a small nonzero , can be replicated by another economy where adjustments occur only at the steady state, i.e. zero . Since the xed cost is exogenous (~ = 0), the degree of aggregate returns to " scale is + (1 ) ]. Because the degree of returns to variety is normalized to 1, an appropriate decrease in the degree of market power by ( 1) is the only modi cation necessary in the new economy without any short-run adjustments. The dynamic behavior of all the aggregate variables, except for the gross output and the number of rms, is the same as that of the original model. More aggregate data may solve the identi cation problem. Data on the xed cost or the number of rms would be helpful. However, it is not likely that we can get measures of these two variables, consistent with this paper. Now suppose that we have additional data on the rental rate or the wage and that the model also includes the appropriate input-price equation, Eq. (26) or (27). The new model identi es another linear combination of parameters, (~ + (1 ") ) + (1 ) ].21 This " ~ is di erent from the degree of aggregate returns and so the model identi es two parameters, unless adjustments are instantaneous. For example, the two free parameters estimated in Kim (1996) are and , both of which are identi ed for the following reason. Since adjustments are assumed to occur only at the steady state, the degree of aggregate returns is and the dynamics of input prices are governed by . Additional data on the interest rate provide information on the rental rate and so is also identi ed. Considering the identi cation problems, one may wonder why we care about various increasing returns separately. This question will be answered when the market economy is compared with a social planner. We will show that di erent increasing returns have di erent normative implications. Before comparing with a social planner, we review the literature using disaggregate data to identify some parameters of the model. Recall that the aggregate economy has been derived from a rm's problem. If we interpret a rm as a particular sector of the economy, the data disaggregated to the sectoral level have implications on the state of the economy. Two related literatures apply this interpretation. Here, we evaluate the implications of these literatures by
Note that the replication of Rotemberg and Woodford (1995) in the previous paragraph does not change input-price dynamics. Since both and are equal to 1, the change in does not change the dynamics.
21

3.2 Interpretation of Disaggregate Data

21

using our general speci cation of increasing returns in a monopolistic-competition model. Firm's rst order conditions are the starting point of testing the joint hypothesis of perfectly competitive markets and constant returns to scale, as in Hall (1996, 1998, 1990). However, compared with the model in this paper, they do not incorporate a zero-pro t condition. Since it is assumed that there are no xed costs, there is no di erence between the structural production function and its reduced form. The log-linearized production function is: The analysis of Hall (1986, 1988, 1990) is based on the measurement of productivity growth in Solow (1957). Calculation of the Solow residual requires the revenue share: Without zero-pro t conditions or xed costs, the revenue share is equal to i (1 Therefore, the Solow residual is:

dyit = yit

i

dkit + (1 i kit

! dlit + dAit : i) lit Ait

W sR = p tylit :
it it

(28)
i ).

This is the basis of testing for the joint hypothesis of perfect competition ( = 1) and constant returns to scale ( i = 1). Hall (1990) proposes a way to di erentiate diminishing marginal cost from market power. He de nes a new share as follows: This is called the cost share, since it is the share of labor input in total cost, rather than in total revenue as in Eq. (28). Note that it is equal to the share parameter, (1 i), regardless of increasing returns and market power. The cost-based residual is as follows: Unlike the Solow residual, the cost-based residual can provide information only on increasing returns.22 This algebra lets Hall (1990) conclude that his evidence points in the direction of increasing returns, presumably coupled with market power.
The similar results for the Solow residual and the cost-based residual support the absence of pro ts in Hall (1990). Note that a zero-pro t condition is not imposed.
22

dyit s dlit (1 s ) dkit = ( R yit R lit kit

1)sR

dlit lit

! dkit + ( kit

i

1) dkit + dAit : k A
it it

sC = Z k Wt lit l : +W
t it t it

dyit s dlit (1 s ) dkit = ( C yit C lit kit

i

! dlit + (1 s ) dkit + dAit : 1) sC l C kit Ait it

22

However, this di erence between the Solow residual and the cost-based residual disappears under the speci cation of this paper which incorporates xed costs and zero-pro t conditions. First order conditions imply that the steady state of the revenue share is the share parameter. The cost share is the share parameter regardless of the speci cation. Considering that the xed cost does not adjust instantaneously, the rm-level reduced-form production function is log-linearized as follows:

dyit = "e + (1 ~ i yit

"e) ] i dkit + (1 ~ kit

! dlit + error; i) lit

(29)

where e represents the elasticity of aggregate inputs with respect to rm inputs.23 The formula for both the Solow residual and the cost-based residual is:

! dyit s dlit (1 s) dkit = "e + (1 "e) 1] s dlit + (1 s) dkit + error; ~ i ~ yit lit kit lit kit where s represents both sR and sC . Since the revenue share is equal to the share

parameter only at the steady state, this formula holds only approximately for the cost-based residual. Since the cost-based residual provides the same information as the Solow residual, the way proposed in Hall (1990) cannot identify the degree of diminishing marginal cost separately from that of market power. While Hall (1986, 1988, 1980) incorporates rst order conditions in his analysis, others study increasing returns directly by regressing outputs on inputs, using Eq. (29) as a regression equation.24 It would be realistic to assume that "e is close ~ to zero, since all three terms are likely to be small. Furthermore, this assumption is true for all existing speci cations summarized in Table 1. Under this assumption, the literature can give information only on the market structure. In equilibrium, the degree of market power is a mixture of two sources of increasing returns at a rm level: diminishing marginal cost and the presence of xed costs. The two sources cannot be separately identi ed by production function regressions which use a reduced form. However, most DSGE papers arguing for or against the existence of indeterminacy draw information on diminishing marginal cost from this literature. This is valid only when there is no xed cost in the economy. We can discuss the existence of indeterminacy only after identifying the degree of diminishing marginal cost separately from that of market power. Furthermore, as we will show shortly, this separate identi cation is important from a welfare perspective.
23 24

dKt lit The elasticity is: e = dkit kit + (1 ) dLt Lt . Kt dlit For a detailed discussion and the references, see Basu and Fernald (1997).

23

A basic issue of welfare economics is whether a market solution will yield the social optimum or not. Unlike the case of monopolistic competition, the zero-pro t condition is not binding for a social planner. So the planner's problem is static in the sense that there is no concern about the speed of adjustments. For ease of comparison, adjustments to zero pro ts are assumed to be instantaneous in the economy of monopolistic competition, i.e. = 1, and so time subscripts are suppressed. The behavior of the market economy is summarized as follows:
market

3.3 Comparison with a Social Planner

= = =

1 1 1

!"

N market Y market

!1

K L1
"

"

;
1 "

!(1 ")(

K L1
1)

;
" +(1 ")

K L1

;

where the level parameter, , is normalized to 1 in the original equations, Eq. (14), (15) and (16). The superscripts of `market' denote the market economy of monopolistic competition. Note that the planner's optimization problem with respect to the xed cost is not well-de ned. For the planner, it is optimal to decrease the xed cost as close to zero as possible. So the welfare issues are considered only under the assumption that the social planner is not allowed to control the xed cost. Given a process of the xed cost, the number of rms in an economy is chosen by the social planner whose objective is: max N ( N

h

)

K L1

N

market

i

:

Due to the partial-equilibrium setup of this paper, it is natural for the social planner to consider aggregate capital and labor as exogenous.25 Note that the market structure a ects the planner's problem only through the exogenous process of the xed cost. For this problem to be well de ned, we assume that > . Under this assumption, the number of rms has two o setting e ects on the aggregate output. An increase in the number of rms increases both aggregate gross output and aggregate xed cost.
This problem is di erent from that of Dixit and Stiglitz (1977) in that the cost function is parameterized and that the utility function is not introduced. This makes it easy to consider monopolistic competition and increasing returns simultaneously.
25

24

The solution of this problem gives the optimal number of rms and the optimal output:

N planner Y planner

= =

1 1

!1

!( 1)

1

!
1

"

K L1 ; ! "( 1) K L1

1 "

" +(1 ")

:

By construction Y planner Y market, where the equality holds when = : A larger degree of returns to variety is an incentive for the social planner to create more rms, i.e. @@ N planner > 0. Noting the similarity between the planner's economy and the market economy, we have the following relation:

Figure 1 on page 27 compares the monopolistic-competition economy with the social planner's. The horizontal axis is the degree of market power ( ). It is easy to see that government intervention has room for welfare improvement by a ecting the number of rms, unless the degree of market power is equal to the degree of returns to variety. When the two degrees match as in the conventional aggregator speci cation of Eq. (5), the social planner has no room for improvement. With an endogenous number of rms (" = 0), the graph of Yp in the rst plot would be horizontal and tangent to Ym at its maximum, where the economy is located. In other words, an economy with the speci cation of Eq. (5) following Dixit and Stiglitz (1977) is not only the best in the class of monopolistic-competition economies but also as good as the social planner's economy. It has been argued that this optimality is due to the feature of constant elasticity of substitution.26 However, the more relevant reason consists in the speci cation that the degree of returns to variety is equal to the degree of market power.27 Note also that this particular optimality result relies on the partial-equilibrium setup of this paper. A general-equilibrium framework featuring endogenous aggregate inputs may break the optimality of the partial-equilibrium setup. For example, the equality between the outputs per rm does not imply the equality between the numbers of rms in Dixit and Stiglitz (1977), and government expenditure is welfare improving in Devereux, Head and Lapham (1996c).
See Romer (1987) for an example. Benassy (1996) also points out that this match \is purely owing to an accidental implicit choice of" the degree of returns to variety.
26 27

8 planner 9 8 >N > < > < planner > N market = < market > = > () > N planner = N market :N : > ; <N

8 planner market 9 >y > <y < = () > yplanner = ymarket > : yplanner > ymarket ;

9 > = >: ;

25

4 Further research
Based on the critical review of the literature, empirical work to identify the parameters needs to be done. Enough data both at aggregate and disaggregate levels might enable the identi cation of all the parameters in this paper. The issue is how to nd a measure, say of the number of rms, relevant to this paper. As to the model, the speci cation of endogenous xed costs needs further analysis. It is more appealing if we have a model which explains how the xed cost changes endogenously in response to the change in exogenous variables. The literature on R&D with entry and exit can be a starting point. Another topic that deserves attention is to include materials as an input to the technology. Such extension makes the aggregation in an imperfectly competitive economy less straightforward, since it involves value added rather than gross output. Furthermore, it might change some results regarding returns to scale.28

See, for example, Hall (1986, 1988, 1990), Rotemberg and Woodford (1995), and Basu and Fernald (1997).
28

26

Aggregate Output

Yp Ym

γ

εγ+(1−ε)ρ

ρ

Number of Firms

Nm Np

γ

ρ

Output per Firm

yp ym

γ

ρ Degree of Market Power (µ)

Figure 1: Comparison with a Social Planner

27

A External Increasing Returns
Here we suppose that externalities come from inputs, rather than outputs. The technology of the rms without externalities, Eq. (1), is rewritten for ease of comparison:29 y = k l1 : Input externalities have been parameterized in two di erent ways. If average inputs a ect the output of rms, their technology is: y = k l1 k l1 ; where k and l are average inputs and denotes the degree of externalities. Following the same procedure of the aggregation and the zero-pro t condition, the degree of aggregate returns to scale in a static model is " ( + ) + (1 ") ]. Note that ~ ~ externalities via average inputs change the degree of diminishing marginal cost from to ( + ), without any change in the degree of returns to variety. In a dynamic case, since the relative size of xed costs depends on the internal increasing returns only, the e ect of market structure is magni ed by a factor of + . That is, the returns to scale in our most general speci cation is, " ! # + " ( + ) + (1 ") ] + (1 ) ~ ~ : Production externalities may come through total inputs as follows: K L1 ; y = k l1 where K and L denote total capital and labor in the economy. Such externalities magnify not only the degree of diminishing marginal cost but also that of returns to variety by a factor of + . The degree of market power is magni ed, too. The degree of returns to scale is, ! + (~ + (1 ") ) + (1 ) ] : " ~ The above algebra shows that a model with external increasing returns and one with internal increasing returns can replicate the dynamic properties of each other, subject to the following quali cation.30 The degree of returns to variety and the degree of market power should be changed appropriately. For example, if internal returns are replaced with external returns, the degree of market power in a new model should be reduced to ( ), the old degree of market power divided by the degree of internal returns.
29 30

In the Appendix, all subscripts denoting rm and time are omitted. This equivalence is also explained in Benhabib and Farmer (1994).

28

B Input Fixed Cost
Suppose that the rms pay xed costs as part of their inputs as well as their output. Then the technology of the rms is:

y = (k k0) (l l0 )1 " K + (1 K Nk0
) L L Nl0

; #

where k0 and l0 are input xed costs. The aggregate version of the zero-pro t condition is:

N = 1

(K Nk0 ) (L Nl0 )1

:

Unless there is no output xed cost ( = 0), this condition is not analytically tractable. Assuming that there is no output xed cost, the aggregate technology and the zero-pro t condition are simpli ed as follows:

Y = N
=

K + (1 K Nk0

(K Nk0 ) (L Nl0 )1
0

) L LNl :

;

With both input xed costs, the zero-pro t condition has a conceptual problem. Interpreted as an equation determining N , it produces two discrete solutions. If k0 and l0 are endogenous, it has a continuum of solutions. Therefore, we need to pin down either K K 0 or L L 0 . With the two ratios de ned as RK and RL , the Nk Nl linearized version is:

dY = ( ) dN Y " N ! !# dK (R 1) d(Nk0) + (1 ) R dL (R 1) d(Nl0) ; + RK K L K Nk0 L" L Nl0 # # " dK d(Nk0 ) + (1 )R (R 1) dL d(Nl0 ) : 0 = RK (RK 1) L L K Nk L Nl
0 0

If the input xed costs are exogenous and the number of rms responds instantaneously, the linearized aggregate reduced-form production function is: dY = ( )+ + (1 )(RL RK + 1) RL 1 dK Y RK 1 K +(1 ) ( ) + (1 ) + (RK RL + 1) RK 1 dL ; RL 1 L 29

where = RK (RK RK (RK )1)L (RL 1) . With all the complexity of the functional form, 1)+(1 R the degree of returns to scale is simply . However, we may assume that more unproductive inputs are necessary in a boom. If the aggregate input xed costs are endogenous, the aggregate production function in a reduced form is:

dY = Y

dK + (1 K

! dL : )L

The degree of returns to scale is . In a dynamic case when the adjustments occur only at the steady state, the linearized aggregate reduced-form production function is:

dY = Y

RK dK + (1 K

! dL : )RL L

The zero-pro t condition implies that the degree of returns to scale is . The model of Yun (1996) introduces only labor xed cost, i.e. k0 = 0, and so magni es only labor uctuations in output dynamics. The bottom line of this appendix is that the model with both input xed costs has richer dynamics in the sense that capital and labor uctuations may be magni ed to a di erent degree. However, this comes at a cost of one additional restriction. Three extreme cases show that the degree of returns to scale is the same as that of the model with only the output xed cost.

30

C Cost Minimization
The framework of cost minimization has been used widely, for example in Hornstein (1993) and Devereux, Head and Lapham (1996a). The cost function of a rm is de ned with rescaled input prices: 1 c (y) = min Zk + (1 )Wl s.t. k l1 = (y + ) ; (k;l) where the general price level, P , is normalized to 1. The rst order conditions produce the conditional demands:

k = l =
So the cost function is:

W Z Z W

1

(y + ) ;
1

(y + ) :
1 1

where mc = Z W 1 1 (y + ) 1 : The marginal cost is decreasing if > 1. However, the second-order su cient condition is satis ed in equilibrium, since xed cost is assumed to be positive. The maximized pro t is:
1

Pro t maximization subject to the output demand of the aggregator, Eq. (8), implies the constant markup rule: p = ; mc

c (y ) = Z W 1 (y + ) :

=Z

W1

(y + )

1

1

"

y (y + ) :

#

Therefore, the condition of zero pro ts at each period results in the following relation: ! = 1 y; which involves output rather than inputs as in Eqs. (4) and (11). This relation shows that output per rm is proportional to the xed cost, whether it is exogenous or not. Interpreting the zero-pro t condition as follows:

N =
=

"

"
31

! #(1 ) 1 Y ; ! # 1 Y ;

we transform the conditional demands and the cost function into an aggregate version as follows:

K = L =
where =
1

W Z Z W

1

Y Y

+1

; Y
+1

(30) (31)

+1

; ;

C (Y ; N (Y ); (N )) = Z W 1 p = ; MC

( ) 1 1 . The constant markup rule is expressed as follows: ( 1)(1 )( ) (32)
1
1

1 ( ) Y . MC is not where p = (1 ) 1 ( 1) the aggregate marginal cost but just a name for the aggregate counterpart of the @ rm's marginal cost. Note that the inverse of the markup is equal to @Y C (Y ; N; ) d evaluated at the zero-pro t condition, not to dY C (Y ; N (Y ); (Y )). Eqs. (30){(32) are a complete description of the economy. Solving for Y as a function of K and L, we have:

h

i(1 )( 1 ) 1 Y and MC = Z W 1

Y=

1K

L1

(

+1

) 1:

The degree of returns to scale is the weighted harmonic average of the degree of diminishing marginal cost and that of returns to variety. The input prices are:

This shows that our results on input prices in a static model are robust to the choice of the framework.

Y Z = K; W = Y: L

32

D Fixed Cost Externalities
In the static case, both the xed cost and the number of rms are endogenous and so the degree of returns to scale is a convex combination of the degree of diminishing marginal cost and that of returns to variety. This appendix shows that the static model is equivalent to two models where the xed cost is exogenous and only the number of rms is endogenous, i.e. " = 0. That is, aggregate output and the number of rms follow Eqs. (10) and (12), rewritten here for convenience. N ; Y = N ( ) K L1

0 1 N = @

11 A K L1 :

Instead of endogenous xed costs, each model introduces an assumption about xed cost externalities. The new parameters, and , represent the external diseconomies such as congestion e ects. First, external diseconomies are assumed to a ect the xed cost as follows: = 1Y ; (33) where 2 0; 1]. An increase in aggregate output would increase the xed cost per rm, less than proportionately. Combining Eq. (33) with the above two equations, the equilibrium aggregate output is: +1 ) 1 Y = K L1 ( ;
1 ( ). The degree of returns to scale is a harmonic where = 1 ( ) 1 average of the degree of diminishing marginal cost and that of returns to variety. Alternatively, negative externalities is assumed with respect to the number of rms as follows: = 2N ; where 2 0; 1). An increase in the number of rms would increase the xed cost per rm, possibly more than proportionately. The same algebra shows that the equilibrium output is: + Y = K L1 ( + ) ( + ) ] ;

(+) 1 + + where = 2 ( + ) 1 1 . As the degree of external2 ity ( ) grows from 0 to 1, the degree of returns to scale moves from the degree of returns to variety ( ) to the degree of diminishing marginal cost ( ). 33

References
1] Ambler, Steve and Emanuela Cardia (1996), \The Cyclical Behavior of Wages and Pro ts under Imperfect Competition," mimeo, University of Montreal. 2] d'Aspremont, Claude, Rodolphe Dos Santos Ferreira, and Louis-Andre GerardVaret (1996), \On the Dixit-Stiglitz Model of Monopolistic Competition," American Economic Review, 86(3), 623{629. 3] Basu, Susanto and John G. Fernald (1997), \Returns to Scale in U.S. Production: Estimates and Implications," forthcoming in Journal of Political Economy. 4] Beaudry, Paul and Michael B. Devereux (1995a), \Money and the Real Exchange Rate with Sticky Prices and Increasing Returns," Carnegie-Rochester Conference Series on Public Policy, 43, 55{102. 5] Beaudry, Paul and Michael B. Devereux (1995b), \Monopolistic Competition, Price Setting and the E ects of Real and Monetary Shocks," mimeo, University of British Columbia. 6] Benassy, Jean-Pascal (1991), \Monopolistic Competition," in W. Hidebrand and H. Sonnenschein, eds., Handbook of Mathematical Economics, Volume IV, 1997{2045. 7] Benassy, Jean-Pascal (1996), \Taste for Variety and Optimum Production Patterns in Monopolistic Competition," Economics Letters, 52, 41{47. 8] Benhabib, Jess and Roger E. A. Farmer (1994), \Indeterminacy and Increasing Returns," Journal of Economic Theory, 63, 19{41. 9] Blanchard, Olivier J. and Nobuhiro Kiyotaki (1987), \Monopolistic Competition and the E ects of Aggregate Demand," American Economic Review, 77(4), 647{666. 10] Chatterjee, Satyajit and Russell W. Cooper (1993), \Entry and Exit, Product Variety and the Business Cycle," NBER WP #4562. 11] Devereux, Michael B., Allen C. Head, and Beverly J. Lapham (1996a), \Monopolistic Competition, Increasing Returns, and the E ects of Government Spending," Journal of Money, Credit, and Banking, 28(2), 233{254. 34

12] Devereux, Michael B., Allen C. Head, and Beverly J. Lapham (1996b), \Aggregate Fluctuations with Increasing Returns to Specialization and Scale," Journal of Economic Dynamics and Control, 20, 627{656. 13] Devereux, Michael B., Allen C. Head, and Beverly J. Lapham (1996c), \Government Spending and Welfare with Returns to Specialization," mimeo, University of British Columbia and Queen's University. 14] Dixit, Avinash K. and Joseph Stiglitz (1975), \Monopolistic Competition and Optimum Product Diversity," University of Warwick WP #64. 15] Dixit, Avinash K. and Joseph Stiglitz (1977), \Monopolistic Competition and Optimum Product Diversity," American Economic Review, 67, 297{308. 16] Feenstra, Robert C. (1994), \New Product Varieties and the Measurement of International Prices," American Economic Review, 84(1), 157{177. 17] Hairault, Jean-Olivier and Franck Portier (1993), \Money, New-Keynesian Macroeconomics and the Business Cycle," European Economic Review, 37, 1533{1568. 18] Hall, Robert E. (1986), \Market Structure and Macroeconomic Fluctuations," Brookings Papers on Economic Activity, 2, 285{322. 19] Hall, Robert E. (1988), \The Relation between Price and Marginal Cost in U.S. Industry," Journal of Political Economy, 96, 921{947. 20] Hall, Robert E. (1990), \Invariance Properties of Solow's Productivity Residual," in P. Diamond, ed., Growth, Productivity, Unemployment, Cambridge: MIT Press, 71{112. 21] Hornstein, Andreas (1993), \Monopolistic Competition, Increasing Returns to Scale and the Importance of Productivity Shocks," Journal of Monetary Economics, 31, 299{316. 22] Kim, Jinill (1996), \Monetary Policy in a Stochastic Equilibrium Model with Real and Nominal Rigidities," mimeo, Yale University. 23] Matsuyama, Kiminori (1995), \Complementarities and Cumulative Processes in Models of Monopolistic Competition," Journal of Economic Literature, 33, 701{729. 35

24] Romer, Paul M. (1987), \Growth Based on Increasing Returns Due to Specialization," American Economic Review Papers and Proceedings, 77(2), 56{62. 25] Rotemberg, Julio J. and Michael Woodford (1995), \Dynamic General Equilibrium Models with Imperfectly Competitive Product Markets," in Frontiers of Business Cycle Research, Thomas F. Cooley (ed.), Princeton University Press, Princeton, NJ. 26] Solow, Robert M. (1957), \Technical Change and the Aggregate Production Function," Review of Economics and Statistics, 39, 312{320. 27] Yun, Tack (1996), \Nominal Price Rigidity, Money Supply Endogeneity, and Business Cycles," Journal of Monetary Economics, 37, 345{370.

36